GPT Image 2.5 compared across all three quality tiers with the same prompt

Riccardo Wolf @WolfRiccardoOriginally posted 2026.09.09

Open original on X

I wanted to compare all 3 quality tiers of the new ChatGPT Images 2.5 model — Sunburst (API). Using the exact same prompt, I generated each image in: • High Quality • Medium Quality • Low Quality With Images 2.0, selecting Low Quality caused a noticeable loss in detail and pixel quality. But with Images 2.5, that’s no longer the case. Both Medium and Low Quality produce surprisingly solid results.